基于集成修剪的丁苯橡胶聚合转化率软测量  

Soft-sensing for polymerization conversion rate of styrene butadiene rubber based on ensemble pruning

作　　者：李炜[1] 章寅[1] 倪源 

机构地区：[1]兰州理工大学,兰州730050 [2]兰州石化公司自动化研究院,兰州730060

出　　处：《仪器仪表学报》2011年第1期212-217,共6页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(60964003);甘肃省自然科学基金(096RJZA101)资助项目

摘　　要：针对目前丁苯橡胶聚合转化率难以在线精确测量,不利于指导生产的问题,本文提出一种基于集成修剪的软测量方法用于丁苯橡胶聚合转化率的预测。首先采用bagging方法建立多个LS-SVM弱学习器,然后利用AdaBoost.RT方法对弱学习器进行修剪,最后将修剪出的弱学习器加权输出。该方法克服了集成算法需要存储空间大和预测时间长的缺点,并且在一定程度上改善了最小二乘支持向量机的稀疏性和鲁棒性问题.仿真结果表明,聚合转化率预报绝对误差大于1.5样本的比例小于10%,能够满足实际生产要求,可以作为过程信息用于丁苯橡胶聚合过程的优化控制。Aiming at the problem that the polymerization conversion rate of styrene butadiene rubber is hard to measure accurately and online,which can not be used to direct production,this paper introduces a new soft-sensing method based on ensemble purning to predict the polymerization conversion rate of styrene butadiene rubber.We first construct multiple LS-SVMs as weak learners with bagging method,and then use AdaBoost.RT algorithm to prune the weak learners.The final output of the pruned weak learners is constructed,which is the weighted combination of the outputs from each selected weak learner.This method overcomes the shortcoming that original ensemble algorithm requires much memory storage and longer prediction time.And to some extent,the proposed method improves the sparseness and robustness issues of LS-SVM.Simulation results indicate that the ratio of the samples for which the absolute error of polymerization conversion rate prediction is greater than 1.5 samples is less than 10%,which meets the practical production requirements;and the method can be used as the process information for the optimal control of styrene butadiene rubber polymerization process.

关 键 词：丁苯橡胶 聚合转化率 集成修剪 BAGGING ADABOOST 最小二乘支持向量机 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]